Abstract:
Building intelligent computer assistants has been a long-cherished goal of
AI. Many intelligent assistant systems were built and fine-tuned to specific
application domains. In this work,
we develop a general model of assistance that combines
three powerful ideas: decision theory, hierarchical task models and
probabilistic relational languages. We use the principles of decision theory
to model the general problem of intelligent assistance. We use a
combination of hierarchical task models and probabilistic relational
languages to specify prior knowledge of the computer assistant. The
assistant exploits its prior knowledge to infer the user's
goals and takes actions to assist the user. We evaluate the decision
theoretic assistance model in three different domains including a real-world domain
to demonstrate its generality. We show through experiments that both the
hierarchical structure of the goals and the parameter sharing facilitated by
relational models significantly improve the learning speed of the agent.
Finally, we present
the results of deploying our relational hierarchical model in a real-world
activity recognition task.
